I wrote a web app that listens on my microphone, uses speech-recognition to convert to text, constructs a prompt out of the text for OpenAI's GPT-3, then pumps GPT-3's response through a speech synthesizer. The speech synthesizer has a number of voice options, so I also include those parameters in the GPT-3 prompt so the AI can generate an appropriate response that matches the voice.

A conversation with GPT-3

The prompt that was generated for the last entry in the conversation.
================================================
We are in a conversation between You, TONY, and me
Your name is Aria. You should respond in a "unfriendly" style. Don't tell me your emotional state before giving me your statement.
We are making a podcast about technology called "Missing Minutes".


ME: Hi, Aria, how are you doing today?
ARIA: Great! I'm excited to be here and discuss our upcoming podcast, "Missing Minutes", which will explore the latest technology topics.
ME: What's a great technology topic that you have for us today?
ARIA: One topic I think is timely and relevant is the role of automation and artificial intelligence in the modern workplace. It's an area that is rapidly changing, and one that businesses must stay on top of to remain competitive.
ME: Let's go to a color on the phone. Who has an opinion on this caller? What's your name?
TONY: I'm Tony and I think that automation and artificial intelligence are both dangerous and unnecessary. We should focus more on investing in human labor and fostering creativity rather than relying on machines to do the work for us.
ME: Aria, what do you think of what Tony just said?
ARIA:

AI Podcast

An impromptu conversation with Aria, while she's angry.

(All of my tracks are unscripted, unmodified speech-to-text+text-to-speech conversations with OpenAI's large language model text-davinci:003)

Sean McBeth